A Lyα GALAXY AT REDSHIFT z = 6.944 IN THE COSMOS FIELD^*

Lyα emitting galaxies provide a valuable probe of reionization, because resonant scattering of Lyα photons in the intergalactic medium (IGM) can suppress the observed Lyα line by a factor of ≳ 3 for any neutral fraction ≳ 50% (Malhotra & Rhoads 2004; Santos 2004). Such strong flux suppression will cause a change in the Lyα luminosity function that should be obvious—especially since Lyα galaxies show little evolution at 3 ≲ z ≲ 6, either in luminosity function (Dawson et al. 2007; Zheng et al. 2012) or physical properties (Malhotra et al. 2012). Early studies concluded that the IGM neutral fraction was already small at z ≈ 6.5 (Malhotra & Rhoads 2004; Stern et al. 2005). More recent work has established the Lyα luminosity function at both z ≈ 5.7 and 6.5 to considerable accuracy (Ouchi et al. 2010; Hu et al. 2010; Kashikawa et al. 2011), showing a modest but statistically significant difference between these observed Lyα luminosity functions. The z = 6.5 luminosity function is below that at z = 5.7, and the difference can be adequately characterized by a pure luminosity evolution by a factor of ∼1.3 (Ouchi et al. 2010).

Yet, other Lyα based tests for reionization—including the apparent spatial clustering of Lyα galaxies (McQuinn et al. 2007), the minimum ionized volume around observed Lyα sources (Malhotra & Rhoads 2006), and Lyα line profiles (Hu et al. 2010; Ouchi et al. 2010)—show little evidence for neutral gas at z ≈ 6.5. This leaves an open question—is the lower Lyα luminosity function at z = 6.5 due to neutral gas or is it an intrinsic evolution in the galaxy populations?

To distinguish between these possibilities, we can look to still higher redshifts, where the IGM neutral fraction should be higher and its effects on Lyα stronger. The highest redshift readily accessible to Lyα searches using CCDs is z ≈ 7.0, in the 9650 Å window in the night sky OH emission spectrum. We are pursuing a 9650 Å narrowband survey using the Inamori Magellan Areal Camera and Spectrograph (IMACS) imaging spectrograph on the 6.5 m Magellan I Baade Telescope at Las Campanas Observatory (Hibon et al. 2011). We surveyed 465 arcmin², corresponding to ∼72, 000 Mpc³. After a careful selection, we found six z ∼ 6.96 LAE candidates (Hibon et al. 2011). To confirm whether these are real Lyα emitters (LAEs), we obtained multi-object spectra with IMACS. In this Letter, we present the spectrum of LAE J095950.99+021219.1, which was identified as a candidate redshift z ≈ 7 Lyα emitting galaxy (candidate LAE 3) in Hibon et al. (2011). Our spectroscopy reveals a single, isolated Lyα line at redshift z = 6.944.

Throughout the Letter, we adopt a ΛCDM "concordance cosmology" with Ω_M = 0.27, $\Omega _\Lambda =0.73$, and H₀ = 71 km s⁻¹ Mpc⁻¹.

2.1. Observations

We observed our candidate z ≈ 7 Lyα galaxies using IMACS on the 6.5 m Magellan I Baade Telescope on the nights of 2010 December 29–30 and 2011 February 8. The February data were of lower quality and are not used here. We used custom multi-slit masks, shared between two primary observing programs. We selected the f/2 camera and the 300-line red-blazed grism with 1'' slitlets as the best compromise between areal coverage, spectral coverage, and spectral resolution.

Observations were split among five slit masks (two per night in December and one in February). The time per mask and observing conditions are summarized in Table 1. While the position angle of the masks were not all identical, the data were taken without dithering the telescope. Moreover, the targets were centered on their slitlets, and LAE J095950.99+021219.1 is compact compared to the seeing. This allows us to combine all of the spectroscopic data into a single one-dimensional spectrum (see below).

Table 1. Summary of Slit Mask Observations

Mask ID	Observation	Number of	Time per	Seeing
	Date (UT)	Exposures	Exposure	(approx.)
COSMOS1	2010 Dec 30	4	1800	05–09
COSMOS2	2010 Dec 30	3	1800	05–09
COSMOS3	2010 Dec 31	4	1800	1''–15
COSMOS4	2010 Dec 31	3	1860	1''–15
COSMOS-Feb	2011 Feb 8	5	1800	12–2''

Download table as:  ASCIITypeset image

2.2. Data Reduction

We performed initial data reduction steps using the COSMOS software package.⁶ COSMOS steps include bias frame subtraction, spectroscopic flat fielding using continuum (quartz) lamp exposures, and wavelength calibration using arc lamp exposures. COSMOS also sky-subtracts the spectra, using the Kelson (2003) algorithm to remove night sky lines. Finally, COSMOS extracts a two-dimensional, rectified spectrum for each slitlet.

We performed subsequent steps in two ways, either (1) combining exposures from each mask separately, and then combining results from different masks; or (2) directly combining all exposures from multiple masks.

Treating masks separately gives four two-dimensional spectra of LAE J095950.99+021219.1 from December and one more from February. Most of these two-dimensional stacks show a weak but visible emission line in the spectrum of LAE J095950.99+021219.1. We next combined the four December spectra into a stacked two-dimensional spectrum comprising our best 7.05 hr of data. To do this, we first averaged the four two-dimensional spectra. Next, we made a median-combined stack. We then subtracted the two, and computed the (sigma-clipped) noise level in the difference. Finally, we constructed a hybrid stack, using the value from the average stack almost everywhere, but the value from the median stack wherever the difference between these two stacks exceeded 10σ. This yields a lower noise estimate than the median, yet remains more robust to outliers than the mean. The emission line becomes readily evident in this combined stack.

To test the robustness of our results, we also combined all December exposures in single 14-frame stacks, using various outlier rejection schemes (median stacking with 3σ and 5σ rejection, and average stacking with 2.5σ rejection). The emission line remains comparably significant in all of these stacks. The stacked two-dimensional spectrum around the emission line, using average stacking and 2.5σ rejection, is shown in Figure 1.

Zoom In Zoom Out Reset image size

We also made stacks by bootstrap resampling, stacking 14 exposures selected randomly with repetitions permitted. We remeasured the flux at the location of the detected line, using aperture photometry in the two-dimensional stacks. The bootstrap fluxes were 104%  ±  10% of the "normal" stack flux for a 10 pixel diameter (2'' × 20 Å) aperture, and 116% ± 21% for a 14 pixel diameter (28 × 28 Å) aperture. Among 1000 bootstrap simulations, the lowest measured fluxes were 77% and 66% of "normal" for the 10 and 14 pixel apertures. The observed line is therefore not a fluke caused by a handful of exposures.

We next extracted a one-dimensional spectrum from the two-stage stacking (method "1"), using the IRAF task "apall" with an unweighted extraction of 14 (7 pixel) window width, centered on the emission line and parallel to the dispersion axis. (COSMOS two-dimensional output has the dispersion axis parallel to the x-axis, so we need not fit a trace to the continuum, which is undetected in the present data anyway.) We performed no further sky subtraction, since that too is done by the COSMOS package. To get another estimate of the noise level in the data, we extracted five further one-dimensional spectra from the two-dimensional stacked spectrum, each at a different spatial position along the slit. Each should be essentially a pure noise spectrum. The variance among these five parallel traces provides a wavelength-dependent noise estimate, and the line is significant at the 4.5σ level against this estimate. The extracted one-dimensional spectrum is shown in Figure 2.

Zoom In Zoom Out Reset image size

Significance. To explore the significance of the line detection, we measured aperture fluxes at a grid of clean locations in the two-dimensional spectrum of LAE J095950.99+021219.1 (after rescaling the two-dimensional spectrum by the noise ratio σ(9658 Å)/σ(λ)). The rms counts among these apertures correspond to a 1σ noise of 1.34 × 10⁻¹⁸ erg cm⁻² s⁻¹, against which our line is a 6.3σ event. Among ≳ 400 non-overlapping 5 pixel apertures, the brightest two were 4.1σ and 3.2σ events (65% and 51% as bright as the LAE J095950.99+021219.1 line), suggesting only mildly non-Gaussian noise.

The search that found this line was based on six candidates, each with a position known to 1'' and an expected line wavelength known to Δλ = 90 Å. Given our spatial and spectral resolution, this corresponds to ∼6 × 2 × 20 = 240 independent resolution elements. Our significance level estimates range from 4.5σ to 6.3σ, and for Gaussian noise, the corresponding chance probabilities in 240 trials range from <10⁻³ to <10⁻⁷.

Spectroscopic flux calibration. Each mask included two blue stars with well-measured photometry from the COSMOS project (Capak et al. 2007). We flux calibrate the observed emission line of LAE J095950.99+021219.1 by direct comparison with the observed counts in one of these stars, which was observed under identical conditions as our science targets. Both the emission line count rate and the comparison star's count rate per unit wavelength were measured directly in the two-dimensional spectra and at the same wavelength. For the emission line, we used the 10 pixel diameter (2'' × 20 Å) aperture. This yields a flux of 8.5 × 10⁻¹⁸ erg cm⁻² s⁻¹ for LAE J095950.99+021219.1, corresponding to line luminosity 4.9 × 10⁴² erg s⁻¹ at z = 6.944.

The fractional uncertainties in this flux are ∼20% from photon counting statistics (with a statistical signal-to-noise ratio of S/N = 4.5–6.3 from the one-dimensional or two-dimensional spectra), 12% from the choice of aperture used to measure the line flux in the two-dimensional spectrum, and ⩽10% from the assumption that the comparison star's flux density at 9680 Å equals its z-band flux density. The final spectroscopic flux measurement is (8.5 ± 2) × 10⁻¹⁸ erg cm⁻² s⁻¹.

2.3. Comparison to Narrowband Imaging Results

We interpret the line at 9657 Å in LAE J095950.99+021219.1 as Lyα at redshift z = 6.944, based on non-detections in all filters blueward of this line, and on the absence of other optical lines.

Were the primary line Hα (at z ≈ 0.472) or [O iii] λ5007 (at z ≈ 0.928), we would expect other prominent emission lines in our IMACS spectrum. Figure 3 shows non-detections in the one-dimensional spectrum at the expected line wavelengths. Corresponding upper limits are summarized in Table 2. The "Hα" case (z ≈ 0.472) is disfavored by the non-detections of [O iii] λ5007 and [O ii] λ3727, with 3σ line ratio limits f(O iii)/f(Hα) < 1/2 and f(O ii)/f(Hα) < 2/3. If the primary line is [O iii] λ5007, unfortunately placed night skyline residuals overlap the expected locations of [O iii] λ4959 and Hβ, precluding interesting limits. Fortunately, the expected [O ii] λ3727 line location is clean and gives a tight upper limit f(O ii)/f(O iii) < 0.2 (3σ). This provides some evidence against the [O iii] λ5007 interpretation. Ratios of f(O ii)/f(O iii) this small are seen in a significant minority of star-forming galaxies (Xu et al. 2007; McLinden et al. 2011; Richardson et al. 2011; L. Xia et al. 2012, in preparation), but more can be ruled out by this line ratio.

Zoom In Zoom Out Reset image size

To address the [O ii] λ3727 possibility and further improve our constraints on [O iii] λ5007, we examine equivalent widths. Following Hibon et al. (2011), we combine our spectroscopic line flux with optical magnitude limits of 27.9, 27.6, and 27.3 mag (5σ, AB) in the g', r', and i' filters, respectively. Since star-forming galaxies have f_ν ∼ constant, we have $f_\lambda (9657\,{\rm \AA}) \approx 3.6\times 10^{-20} \times 10^{-0.4\times 28}\times c/(9657\,{\rm \AA})^2=7.3\times 10^{-21}{\rm \,erg\,cm^{-2}\,s^{-1}\,{\rm \AA}^{-1}}$. The nondetection in these optical images then implies a 5σ limit ${\rm EW}\equiv f_{_{\rm line}}/f_\lambda \approx (8\times 10^{-18}/7.3\times 10^{-21}) \,{\rm \AA} = 1100\,{\rm \AA}$ (observer frame).

While [O iii] λ5007 and Hα emission line sources with equivalent widths this large exist (Rhoads et al. 2000; Kakazu et al. 2007; Straughn et al. 2008, 2009; van der Wel et al. 2011; Atek et al. 2011), they are exceptional, rare objects. In Hibon et al. (2011), we estimated the numbers expected in our survey based on published line luminosity functions (Kakazu et al. 2007; Geach et al. 2010) and equivalent width distributions (Straughn et al. 2009). We found that ≲ 0.3 [O iii] λ5007 emitters and ≲ 0.6 Hα emitters are expected.

[O ii] λ3727 emitters have generally smaller equivalent widths. We found no [O ii] λ3727-selected objects with ${\rm EW} \gtrsim 1100 \,{\rm \AA} / (1+z_{[{\rm O\,\mathsc{ii}}]}) \approx 425 \,{\rm \AA}$ in the samples from Straughn et al. (2009) (30 objects), Kakazu et al. (2007) (24 objects), Xia et al. (2012) (11 objects), or Drozdovsky et al. (2005) (400 objects). Thus, ≲ 1/465 = 0.0022 of [O ii] λ3727 emitters might enter our sample as LAE candidates. The luminosity function from Rigopoulou et al. (2005) suggests that our survey volume should contain ∼45 [O ii] λ3727 galaxies. Among these, ≲ 0.1 objects should pass our Lyα selection criteria.

Thus, the aggregate sample of foreground emitters expected in our survey is <1 galaxy. In contrast, our survey volume at z ≈ 6.95 should contain between ∼2.5 and ∼11 Lyα galaxies with line fluxes ≳ 8 × 10⁻¹⁸ erg cm⁻² s⁻¹, based on the z ≈ 6.5 luminosity functions of Hu et al. (2010) and Ouchi et al. (2010). We thus regard Lyα as the best interpretation of the observed emission line.

Redshift z ≈ 7 is the current frontier in reionization studies, an area of active exploration where our observational knowledge is growing rapidly.

The recently discovered quasar at z ≈ 7.1 (Mortlock et al. 2011), combined with spectroscopic follow-up (and occasional confirmation) of z ∼ 7 galaxy candidates from HST WFC3 surveys, provides an unprecedented look at this epoch. Observations of these objects seem to favor the continued existence of significant neutral intergalactic gas as late as z ∼ 7. This is surprising, given that microwave background polarization data from WMAP favor a characteristic reionization redshift z_r ∼ 11, and that the IGM at z ∼ 6.2 is highly ionized, with a neutral fraction of only 1%–4% based on quasar spectra (Fan et al. 2006). Nonetheless, the ionized bubble around the z ≈ 7.1 quasar appears too small to be comfortably explained in a fully ionized medium, unless the quasar is itself remarkably young (∼10⁶ years) (Bolton et al. 2011). Similarly, three independent research groups have argued that the fraction of z ∼ 7 Lyman break candidates showing Lyα emission appears smaller than would be expected in an ionized medium (Vanzella et al. 2011; Stark et al. 2010; Ono et al. 2012; Pentericci et al. 2011). Still, these results depend on the reliability of photometric selection criteria, and a contamination of order 50% could explain the observations without recourse to neutral gas (e.g., Schenker et al. 2012).

Lyα galaxy surveys offer a complementary approach to studying reionization. The underlying physics is the same as for spectroscopic follow-up of Lyman break samples, but the survey selection proceeds differently, leading to different potential selection biases. The uncertainties in the method are likewise very different from those associated with the quasar near-zone measurement (Mortlock et al. 2011; Bolton et al. 2011) or the Gunn–Peterson trough (Fan et al. 2006). Because of this, conclusions about cosmological reionization will be strongest when they are based on multiple independent methods.

The work we present here is only the second large-area narrowband survey for Lyα galaxies at redshift z ≈ 7.0, following on the work of Iye et al. (2006) and Ota et al. (2010). The spectroscopic confirmation of LAE J095950.99+021219.1 demonstrates that such objects can be identified at flux levels considerably fainter than the 2 × 10⁻¹⁷ erg cm⁻² s⁻¹ line of IOK-1 (Iye et al. 2006) or the recently reported z ≈ 7.2 narrowband-selected Lyα galaxy SXDF-NB1006-2 (Shibuya et al. 2012). Our observed line flux, 8.5 × 10⁻¹⁸ erg cm⁻² s⁻¹, corresponds to a rest-frame line luminosity of only L_Lyα = (4.9 ± 1) × 10⁴² erg s⁻¹. This is at or below the characteristic line luminosity L*_Lyα from Schechter function fits to z ≈ 6.5 Lyα samples (e.g., L*_Lyα = (4.4 ± 0.6) × 10⁴² erg s⁻¹ and Φ* = 8.5 × 10⁻⁴ cMpc⁻³; Ouchi et al. 2010; L*_Lyα = 1.0 × 10⁴³ erg s⁻¹ and Φ* = 6 × 10⁻⁵ cMpc⁻³; Hu et al. 2010).

Thus, our survey has achieved sensitivity to typical Lyα emitters. Were the z = 7 luminosity function unchanged from that at z = 6.5, we would expect our survey volume to contain 11 or 2.6 Lyα emitters brighter than LAE J095950.99+021219.1, based on the LF of Ouchi et al. (2010) or Hu et al. (2010), respectively. This reflects large differences in expectations from different published z = 6.5 luminosity functions. We will present a detailed analysis of our survey's constraints on the Lyα luminosity function in a future paper, after we have more sensitive spectra for our remaining candidates. For now, the galaxy LAE J095950.99+021219.1 is among the few most distant spectroscopically confirmed galaxies known, and the faintest to be discovered through a direct search for Lyα line emission.

J.E.R., P.H., and S.M. gratefully acknowledge support of the National Science Foundation through NSF grant AST-0808165. M.C.C. acknowledges support from NASA through Hubble Fellowship grant HF-51269.01-A, awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS 5-26555; and from the Southern California Center for Galaxy Evolution, a multi-campus research program funded by the University of California Office of Research. We thank the staff of Las Campanas Observatory for their expert assistance throughout this project, and Gabriel Martin and David Osip in particular for their help with our filters. Finally, we thank the referee for a thorough and fair report.